When Work Runs on a Meter, Who Pays for Everything Else?

Every rent check, mortgage payment and grocery bill in America sits on one old bargain. A person shows up, sells scarce hours to an employer, and takes home a paycheck. Based on Bureau of Economic Analysis figures for the third quarter of 2026, US employee pay and benefits run at a $16.5 trillion annual rate. That's half of a $33 trillion economy.

My argument is that this bargain is starting to break in a specific way. Work is beginning to sell the way electricity sells, off a meter, by the unit, from a plant you never see. And once labor turns into a utility, the hardest questions land on the wage bill, the tax base, and who gets a check when the paycheck shrinks.

What one human hour costs

Right now 163.6 million Americans work. The average private sector worker earns $38 an hour in wages, and once benefits and payroll taxes are added, one hour cost private employers $46.89 in June 2026, based on BLS figures.

That hour is a scarce input that gets more expensive as demand rises, because people need sleep, vacation, healthcare and management. In the third quarter of 2026, median weekly earnings for full-time US workers sat at $1,232, about $64,000 a year. That was up 1.5% while consumer prices climbed 3.5%.

Software agents and humanoid robots run the other way. They get cheaper as they scale, and that pushes the cost of work toward the raw price of silicon and electricity.

Nobody hires the cyclist

Why doesn't anyone pay a neighbor an hourly wage to pedal a bicycle generator in the backyard to keep the refrigerator cold? Because the electric company sells power for about 18 cents per kilowatt hour on a US home bill. You plug in, pay at the end of the month, and never think about the power plant.

Set that next to a person. At the August 2026 US industrial rate of 9.64 cents per kilowatt hour, the roughly $47 a business spends on one employee hour buys hundreds of kilowatt hours of power.

A utility sells a metered flow. You don't buy the station. You buy kilowatt hours, cubic feet of gas, compute tokens and completed tasks. Labor is joining that list, because a company can now buy finished digital work by the task and will soon buy physical work by the hour. Humans stop providing commodity labor and start deploying it.

How a meter takes over

Electricity made this jump over a century ago, and it did it slowly. In the 1880s, Thomas Edison sold isolated generators into single factories and rich city blocks, so every owner kept a private generator in the basement. Then Samuel Insull of Chicago's Commonwealth Edison built central stations on one shared grid. He spread the load across day and night users and accepted state price regulation in exchange for a franchise.

Even then, Stanford economic historian Paul David showed in his 1990 study of the electric dynamo that factory owners spent decades bolting electric motors onto old steam mill layouts before redesigning buildings around the wire. Labor will follow factory math instead of hiring math, and that is where the old market breaks.

The office meter is already running

White collar workers assumed automation would hit warehouses first. But software has no gravity and no metal supply chain, so the meter for office work went live first. Between August 11th and October 21st, 2026, four frontier labs shipped products that crossed from chatbots to autonomous digital workers.

xAI's Grok Bot, in beta since August 11th, runs inside its own cloud computer, and its product page promises agents that keep working 24/7. Meta's Muse launched September 8th, and Meta's Alexandr Wang said on November 10th that Muse now has its own email address for paid users in the US and Canada. OpenAI's Dots, introduced September 29th, connect to over 4,000 apps. OpenAI's help center says they become available by default for eligible enterprise workspaces starting November 18th unless an administrator opts out. Anthropic's Claude Duet, opened as a research preview on October 21st, can use over 6,000 connectors and plugins and keeps working when the user steps away.

Then look at the bill. According to OpenAI's help center, eligible enterprise workspaces pay for dots through flat rates mapped to the type of work, or scaled consumption. So an enterprise no longer pays only per human seat. It buys a metered rate tied to the work the agent finishes. That's a utility tariff for office labor.

The capability underneath is on a fast clock. METR measures how long a software task a frontier agent can finish with 50% reliability, and its January 2026 update put the doubling time at roughly 89 days since 2024. Menlo Ventures' November 2026 report estimates enterprise AI spending jumped from $33 billion in 2025 to $176.8 billion by September 2026. On that path, I expect agents to handle almost any routine digital task by the end of 2027.

Bodies on the same meter

Physical labor is the bigger prize. On the video's estimate it is a market north of $40 trillion a year, roughly 30% of a world economy the IMF puts above $128 trillion. In Tesla's third-quarter 2026 update on October 21st, the company wrote, "Optimus production began in Fremont on our high-volume lines." Figure reported that during an 11-month deployment at BMW's Spartanburg plant, its Figure 02 robots logged over 1,250 hours of runtime, loaded over 90,000 parts and contributed to the production of more than 30,000 BMW X3 vehicles.

Now run the plant manager's math. A full-time worker logs about 1,800 hours a year. A humanoid running 20 hours a day, every day, logs about 7,300. One robot covers the working hours of roughly four people.

Then price the hour. On the video's math, a $20,000 to $30,000 robot, Tesla's stated long-term unit target, spread over 3 years at roughly 7,000 hours a year plus electricity and maintenance, lands around $1.50 to $2 an hour by the end of the decade. Early commercial fleets circulate at $3 to $5 an hour. Compare that with $22 to $28 for US warehouse workers and roughly $40 to $46 for the average fully loaded private employee. You pay the labor company for robot hours the way you pay the electric company for kilowatt hours.

None of this happens overnight. On tape I put meaningful Optimus volume in 2027, with external deliveries targeted toward the end of 2027, starting in factories and warehouses. But the shock starts the moment capital can buy work off a meter instead of hiring a household.

The paycheck slice is shrinking

The government's own data shows the jaws opening. On November 5th, 2026, the Bureau of Labor Statistics reported third quarter nonfarm business output up 3.7% at an annual rate, while hours worked rose just 0.1%, a 3.6% jump in labor productivity. In the same release, labor's share of nonfarm business output fell to 53%, which BLS identified as the lowest level in a series that begins in the first quarter of 1947.

So output grows and the paycheck slice shrinks. The rest flows to the owners of capital, compute and automated systems. An August 2026 VoxEU column from the Centre for Economic Policy Research analyzed Anthropic's economic index across 86 countries. It estimated the labor cost equivalent of observed AI usage at $2.7 trillion a year, 3.4% of GDP, more than doubling in 6 months.

And look at who owns the plant. Based on late October 2026 earnings guidance, four companies, Alphabet, Amazon, Meta and Microsoft, are deploying $720 to $740 billion in a single year on data centers, chips and power. The Menlo Ventures report says Anthropic, OpenAI and Google collect 69 cents of every enterprise AI deployment dollar.

The bottom rung goes first

Farm mechanization took agriculture from about 41% of American workers in 1900, according to the US Department of Agriculture, to under 2% today. The country didn't end up permanently unemployed, because workers moved into factories and offices. This time the machine arrives in the office and the factory at once.

In a November 2025 60 Minutes interview, Anderson Cooper read back a warning from Anthropic CEO Dario Amodei that AI could wipe out half of all entry-level white-collar jobs and spike unemployment to 10% to 20% in the next one to five years. Amodei said "Yes," and added, "that is the future we could see, if we don't become aware of this problem now." Measured US unemployment in October 2026 was 4.1%.

The fracture shows up first in payroll data. An August 2026 Stanford Digital Economy Lab study of ADP records through June 2026 found no mass firing wave. The pain sits with workers ages 22 to 25 in the most AI-exposed jobs. Their employment fell about 11% between November 2022 and June 2026 and now sits 19% below where it would be if it had kept pace with young workers in less exposed jobs. Companies are keeping senior staff and freezing the bottom rung of hiring while agents do the junior coding, research and drafting.

I expect AI to remove commodity jobs faster than new industries replace them over the next 5 to 10 years. New work in fleet orchestration and high-touch services is real. But it doesn't fix the tax math.

Eighty-six cents of every federal dollar

How does a government collect taxes when the newest 100 million workers don't get a paycheck? Based on Congressional Budget Office and Treasury figures for fiscal year 2026, the federal government collected about $5.4 trillion, spent $7.4 trillion and ran a $2 trillion deficit. Individual income taxes brought in about $2.85 trillion, 52.7% of receipts. Social Security and Medicare payroll taxes added about $1.82 trillion, roughly 33.6%. Together that's about 86% of federal revenue, most of it levied on wages and salaries. Corporate income taxes brought in just $382 billion, 7.1%, down 16% from the year before.

Payroll programs are already strained. The Social Security Administration's June 2026 trustees report release shows the combined trust funds collected $1.32 trillion in net payroll taxes in 2025, spent $1.61 trillion, nearly all of it on benefits, and drew down reserves by $160 billion. Under intermediate projections, the retirement trust fund runs out of reserves in the fourth quarter of 2032, leaving payroll taxes able to cover only 78% of scheduled checks. Medicare's hospital insurance fund follows in the second quarter of 2033, covering 89%.

Now follow one hour. When a company swaps a $46 human hour for a $2 robot hour or a software meter, the worker's income tax and both halves of the payroll tax disappear from that hour. The meter gets paid. The Treasury doesn't.

Gates wanted to tax the robot

Bill Gates spotted this hole in February 2017, in an interview with Quartz: "Right now, the human worker who does, say, $50,000 worth of work in a factory, that income is taxed and you get income tax, social security tax, all those things. If a robot comes in to do the same thing, you'd think that we'd tax the robot at a similar level." He even said governments "ought to be willing to raise the tax level and even slow down the speed of that adoption somewhat."

I think slowing it down is a trap. Say one city, state or country slaps a punitive tax on robots and agents to keep humans pedaling the bike. Businesses in the next jurisdiction will outproduce and underprice them into irrelevance. So the answer has to be letting ordinary people own and draw value from the utility itself.

The last time a utility got cornered

When giant holding companies cornered the American electric grid in the 1920s, the country didn't ban power plants. It built a menu of ownership, from regulated investor-owned utilities to city-owned municipals, rural cooperatives and federal power like the Tennessee Valley Authority. The video lays out three mechanisms for labor.

Checks, and the Alaska problem

The first is a direct income floor, through a negative income tax, universal basic income, or what Elon Musk calls universal high income. At the Saudi US Investment Forum in May 2025, Musk told Bloomberg Television, "I think it's actually going to be universal high income." In an April 16th, 2026 post on X cited by Business Insider, he wrote that universal high income "via checks issued by the Federal government is the best way to deal with unemployment caused by AI."

Checks carry a political risk, and Alaska shows it. The Alaska Permanent Fund held roughly $88 billion in July 2026, and the governor's budget put the full statutory formula dividend at $3,650 per resident. The legislature set the 2026 dividend at $1,000, plus a $200 energy relief payment, for $1,200 paid starting October 1.

Owning a piece of the plant

So the second mechanism is direct citizen ownership, the way Norway's government pension fund global holds roughly 72% of its portfolio in global company shares for the public. In his March 2021 essay "Moore's Law for Everything," OpenAI CEO Sam Altman wrote, "Even more power will shift from labor to capital." He proposed an American equity fund, capitalized by taxing companies above a valuation threshold 2.5% of their market value each year in shares plus 2.5% of privately held land value, paying shares and cash to every adult citizen each year.

The third mechanism puts abundance into people's lives early. A company that wins permits for a massive AI data center can pay residential electricity and water bills in that zip code. Child accounts can be funded at birth with shares of leading AI and robotics companies. Cities can offer free self-driving transit miles and community robot hours as public infrastructure.

Sharing the meter

The rule is simple. Laws can't push the value toward a small number of people, and they can't slow or halt the technology. Nobody is going to uninvent the digital agent or the humanoid robot. Nobody misses the cyclist either. What's left to decide is who owns the outlet. Labor is already moving from a scarce hourly wage to an abundant metered utility. The only question left is how fast and how broadly society shares the meter.